Bake Cosmetics turns a described skin concern into the right active
Turning a described skin concern into the right active
- Client
- Bake Cosmetics
- Platform
- Shopify
- Market
- India
- Window
- 24 Jul to 7 Sep 2026
What Bake Cosmetics has live
Carries the pre-sale chat on a catalog where nearly every product name is a chemical name.
Takes a shopper who says "dark spots" or "whiteheads" and walks them to a specific format and strength.
Handles the order status questions that would otherwise sit in the same queue as pre-sale questions.
Bake Cosmetics sells treatment skincare built on named actives: kojic acid and alpha arbutin for pigmentation, niacinamide and azelaic acid for marks and texture, salicylic, glycolic and lactic acids for exfoliation. It is a catalog organised by ingredient and by concern, not by shade or by finish. A shopper who knows they have dark spots still has to work out whether that means a soap, a serum, a body lotion or a cream. This is a small deployment and the numbers below are stated against a base of several hundred chats across 46 days. It is included because the shape of the problem is unusually clear, not because the volume is large.
Challenge: shoppers describe a symptom, the catalog lists a molecule
Bake's shoppers do not arrive asking for a product. They arrive describing a face. The most common openings across the window were whiteheads and blackheads, acne marks, dark spots and pigmentation, dull skin, and dark underarms.
The catalog answers in a different language. It is indexed by active and by concentration: 2% kojic acid, 10% niacinamide, 8% glycolic acid, 10% AHA with 1% BHA. Both vocabularies are correct and they do not meet without a translation step.
The second theme is patience. Timeline questions ran through the whole window, because actives work slowly and a shopper who expects a week and waits a month is a refund. Order tracking was the single most frequent individual request in the account.
- “When can I expect visible results? 10% AHA and 1% BHA Scrub Treatment Soap”
- “I want to track my order.”
- “Which one should I use for acne marks and dark spots?”
- Timeline answers for exfoliating actives were incomplete, so the results question went unresolved
- Vague openings such as "oily acne" produced a recommendation that did not always match the concern
- Benefit questions were answered without closing the loop on which product to actually buy
Solution: guided discovery in front of the ingredient list
The Website Agent carries the catalog, including what each active does, what strength it comes in, and which format suits body versus face.
AI Quiz and Discovery is the piece that matters here. It takes the shopper's own words, a concern rather than a molecule, and narrows to a format and a strength before recommending anything. 74 chats ran through it in 46 days, the third largest classified intent in the account.
WhatsApp AI absorbs order tracking. Order status, enquiry, cancellation and returns together account for 55 chats, and keeping them off the pre-sale surface leaves the Website Agent doing the job it is good at.
Strategy: how Bake runs it now
Let the shopper describe the problem, not the product. Product Info at 146 chats and Product Discovery at 86 are the two largest intents, and together with quiz chats they make up 74.8% of all classified intent. This is a pre-sale deployment, not a support one.
Close the loop on benefit questions. The clearest recorded gap is an agent that explains what an active does and then stops. Naming the product at the end of that explanation is a configuration change, not a model problem, and it is the highest leverage fix in the account.
Publish the timelines. Results timeline questions recur through the whole window and the answers were flagged as incomplete more than once. A published, per-active timeline would answer them in the chat and earn the search query at the same time.
Chat intent mix, 24 Jul to 7 Sep 2026
| Intent | Share of classified intent |
|---|---|
| Product Info | 35.4% |
| Product Discovery | 20.9% |
| AI Quiz and Discovery | 18.0% |
| Product Browsing | 9.5% |
| Order Tracking | 6.8% |
| Order Enquiry | 3.4% |
| Checkout Enquiry | 2.7% |
| Returns, refunds and cancellations | 3.2% |
| Offers and rewards | 0.2% |
Results, 24 Jul to 7 Sep 2026
| Metric | Result | Note |
|---|---|---|
| Resolved without human handoff | 94.06% | several hundred chats |
| Chat engagement | 14.95% of sessions | several hundred of over 2,000 sessions |
| Chats referencing a product | 42.24% | several hundred |
| Guided discovery chats | 74 | AI Quiz and Discovery |
| Human handoffs | 18 | across 46 days |
| Pre-sale share of classified intent | 74.8% | not a support deployment |
Rates are published; revenue amounts, order counts and average order values are withheld at the brand's discretion. The chat base is several hundred across 46 days and is stated on every figure above so the sample size is never implied to be larger than it is.
Is this replicable for a brand like yours?
The conditions this deployment ran under, so you can judge how close your own store is.
| Dimension | Bake Cosmetics |
|---|---|
| Catalog shape | Indexed by active and concentration, not by shade |
| Channels live | Website Agent, WhatsApp AI |
| Buying decision | Concern to format to strength |
| Shopper vocabulary | Symptoms, not ingredients |
| Discovery surface | AI Quiz and Discovery, 18% of classified intent |
| Device split | Majority mobile |
| Market | India |
Steal this playbook
- 01
Index your catalog the way shoppers describe the problem
Bake's products are named for molecules and its shoppers arrive naming symptoms. The gap between those two vocabularies is the whole job.
- 02
Always name the product at the end of a benefit answer
An explanation without a recommendation is a chat that resolved and did not sell.
- 03
Publish per-active results timelines
Slow-acting actives generate the same patience question forever, and answering it in public earns the query as well as the sale.
- 04
Keep order tracking off the pre-sale surface
It is the single most frequent request here and it has nothing to do with choosing a product.
- 05
State a small base plainly
A 94% resolution rate on several hundred chats is a real result and a modest sample. Saying both is what makes the first number believable.
Questions about this deployment
YourLio transformed our website from a product catalog into a trusted beauty advisor.
More brands running YourLio
All case studies
Published 2025-12-10 · Last updated 2026-09-08 · Window 24 July 2026 to 7 September 2026 (46 days), measured from YourLio daily snapshots. Revenue amounts, order counts and average order values are not published.